About
I am an AI/ML Engineer with 5+ years of experience building production-grade machine learning systems across healthcare imaging and financial services. I specialize in the full ML lifecycle — from data engineering and model development to cloud deployment and real-time monitoring. At Philips Healthcare, I lead computer vision pipelines that process over 10,000 chest X-rays daily, achieving 92% anomaly-detection precision, and integrate LLMs to automate radiology report generation. At HDFC Bank, I built NLP-driven fraud detection systems that analyzed over 5 million daily transactions with 95% recall — helping protect customers at scale. I'm experienced across a broad stack, including PyTorch, TensorFlow, Hugging Face Transformers, LangChain, AWS SageMaker, Azure ML, and Kubernetes, with a strong focus on MLOps practices like CI/CD automation, federated learning, and model explainability. I'm passionate about building AI systems that don't just perform in research settings, but solve real problems in production — reliably, responsibly, and at scale.
Professional Experience
AI/ML Engineer
Philips Healthcare -- Orlando, FL
Dec 2023 -- Present
- --Led end-to-end development of computer vision models using PyTorch and OpenCV for medical
- --image analysis, processing 10K+ X-ray scans daily to detect anomalies with 92% precision. Deployed
- --via AWS SageMaker pipelines, slashing model training time by 40% through hyperparameter tuning
- --with Weights & Biases.
- --Designed and fine-tuned CNN architectures (ResNet, YOLOv8) on custom datasets, achieving 28%
- --accuracy uplift in diagnostic predictions for lung disease detection. Integrated with Kubernetes for
- --scalable inference serving across clinical workflows.
- --Orchestrated data pipelines with Apache Spark and Pandas for preprocessing 500GB+ imaging data,
- --implementing augmentation techniques that improved model robustness by 25%. Monitored
- --performance drift using MLflow for production-grade reliability.
- --Built real-time inference APIs with FastAPI and TensorRT optimization, reducing latency from 2s to
- --150ms on edge devices for radiologist tools. Collaborated with clinicians to iterate models via A/B
- --testing in SageMaker.
- --Implemented federated learning protocols using Flower framework to train models on distributed
- --hospital data without compromising HIPAA compliance. Resulted in 15% better generalization across
- --5 regional sites.
- --Developed custom evaluation metrics and bias detection scripts with scikit-learn, ensuring ethical AI
- --deployment that passed internal audits with zero flags. Automated reporting dashboards in
- --Matplotlib/Seaborn.
- --Led MLOps automation with Docker, Jenkins CI/CD, and GitHub Actions, accelerating release cycles
- --from 4 weeks to 1 week for quarterly model updates. Handled versioning of 20+ artifacts in MLflow.
- --Integrated LLMs (BERT variants via Hugging Face) for automated report generation from image
- --analyses, cutting radiologist review time by 35%. Fine-tuned on domain-specific corpora for clinical
- --accuracy.
- --Conducted code reviews and mentored 4 junior engineers on PyTorch best practices, fostering PEP8-
- --compliant, modular codebases that enhanced team velocity by 20%.
AI/ML Engineer
HDFC Bank -- Bengaluru, India
Dec 2019 -- Nov 2022
- --Directed end-to-end NLP pipeline for fraud detection using Hugging Face Transformers and
- --TensorFlow, analyzing 5M+ daily transactions to flag anomalies with 95% recall. Deployed on Azure
- --ML for real-time scoring, reducing false positives by 30%.
- --Engineered LSTM and Transformer models for time-series forecasting of transaction risks,
- --incorporating Graph Neural Networks (PyG) that predicted fraud rings with 85% accuracy. Scaled via
- --Kafka streams for high-velocity data.
- --Curated and labeled 1M+ transaction datasets with Dask and spaCy, applying SMOTE oversampling
- --to balance classes and boost F1-score by 22%. Automated feature engineering with custom Pandas
- --pipelines.
- --Optimized gradient boosting models (XGBoost, LightGBM via scikit-learn) for credit risk assessment,
- --delivering 18% improvement in AUC-ROC over baselines. Served predictions through Kubeflow
- --pipelines.
- --Developed explainable AI layers with SHAP and LIME, generating compliance reports that satisfied
- --RBI audits and cut investigation times by 40% for flagged cases. Visualized insights in Seaborn
- --dashboards.
- --Implemented ensemble methods combining CNNs for transaction image OCR and RNNs for
- --sequence modeling, achieving 32% fraud loss reduction across 2M accounts. Monitored with
- --Prometheus/Grafana.
- --Automated model retraining workflows using MLflow and Airflow, processing weekly data drifts and
- --maintaining 99.9% uptime during peak banking hours. Handled 50TB+ historical data in Hadoop.
- --Integrated generative AI (GPT-like models via LangChain) for synthetic data generation, expanding
- --training sets by 5x while preserving privacy via differential techniques. Validated with custom
- --metrics.
- --Collaborated in Agile sprints via JIRA, leading module delivery from data ingestion to deployment,
- --resulting in on-time rollout of 3 major fraud systems ahead of schedule by 2 weeks.
Skills
Education
Master of Science in Information Systems
Marist College -- Poughkeepsie, NY
Bachelor of Technology in Mechanical Engineering
SIR C.R.R. College of Engineering -- Eluru, India
Contact
Interested in working together? Feel free to reach out.
Irving , TX